Main idea of this school is to raise awareness and applicability about Deep Learning (DL)
with Computational Intelligence (CI) among the research communities. Traditional learning and computing methods
deal with vast range of applications related to reasoning, decision making, perception building etc. However,
DL with CI deals with dynamical systems more efficiently by embedding and facilitating learning mechanism.
Most of the data obtained in real time environment from various domains i.e. environment, industry, business,
biology involves lots of imprecision and vagueness. To address these issues in various domains of algorithms,
tools or techniques are required which should be adaptive and robust so that they can handle the uncertainty and
dynamic nature of this system in efficient and optimized way.
“CI is a set of biological and linguistic tools and methodologies to address complex real-world problems to which traditional Artificial Intelligence (AI) approaches may not be very effective. CI comprises of concepts and implementations that ensures intelligent behavior in complex and dynamic environment.” According to Robert J. Marks, “Neural Network, Genetic Algorithms, Fuzzy systems, evolutionary programming and artificial life are the building blocks of Computational Intelligence.” Using CI tools, we will be able to build the systems which are prone to adaption, robust across problem domains, apply extrapolated reasoning and behave intelligently in given state. Neural network techniques provide capability of computational adaption. System can improve its parameter without any intervention based on optimizing criteria same as human learning occurs. Fuzzy systems help in defining the system where we have a rough estimate of system requirements. Evolutionary algorithms are good enough to optimize parameters and to select best among given constraints. Synergistic effect of these tools may increase their individual performances and gives better adaptive and reliable system. Knowledge representation, reasoning, information mining, discovery science, web intelligence, semantic web, multi agent systems and designing of products i.e. air conditioners, automobile systems, ABS, cameras, dishwashers, pattern recognition in remote sensing, video games are the major areas where CI can be very helpful.
Advantages of DL and CI over existing deep learning algorithms: One has automatic structure optimization ability where a neural network structure (e.g., the number of layers, the number of units in each layer, the type of an activation function at each unit, etc.) is automatically optimized for a given data set and a given objective by an evolutionary structure learning technique. The other is multi-objective ability where a number of different neural networks are simultaneously obtained under a multi-objective scenario. Various multi-objective formulations can be considered for deep learning. A general formulation is a combination of complexity minimization and accuracy maximization. For detection problems, false positive and false negative can be handled as separate objectives.
The school will bring people working in CI and DL domain to a common platform for generating innovative ideas. The school will also assess the state of the art on what new directions lie open for research in area of CI and DL. In this way, the school will generate exciting new communication across various CI and machine learning disciplines.